We investigate the evolution of Brightest Cluster Galaxies (BCGs) from redshift z~1.6 to z~0. We use the semi-analytic model of Croton et al. (2006) with a new spectro-photometric model based on the Maraston (2005) stellar populations and a new recipe for the dust extinction. We compare the model predictions of the K-band luminosity evolution and the J-K, V-I and I-K colour evolution with a series of datasets, including Collins et al. (Nature, 2009) who argued that semi-analytic models based on the Millennium simulation cannot reproduce the red colours and high luminosity of BCGs at z>1. We show instead that the model is well in range of the observed luminosity and correctly reproduces the colour evolution of BCGs in the whole redshift range up to z~1.6. We argue that the success of the semi-analytic model is in large part due to the implementation of a more sophisticated spectro-photometric model. An analysis of the model BCGs shows an increase in mass by a factor ~2 since z~1, and star formation activity down to low redshifts. While the consensus regarding BCGs is that they are passively evolving, we argue that this conclusion is affected by the degeneracy between star formation history and stellar population models used in SED-fitting, and by the inefficacy of toy-models of passive evolution to capture the complexity of real galaxies, expecially those with rich merger histories like BCGs. Following this argument, we also show that in the semi-analytic model the BCGs show a realistic mix of stellar populations, and that these stellar populations are mostly old. In addition, the age-redshift relation of the model BCGs follows that of the universe, meaning that given their merger history and star formation history, the ageing of BCGs is always dominated by the ageing of their stellar populations. In a LambdaCDM universe, we define such evolution as "passive in the hierarchical sense".